cofounder @onyxbio

Joined June 2013
26 Photos and videos
top notch investors and even better human beings
Apr 26
[brief update] we have a new set of funds for @haystackvc announcing haystack 8 plus more, but it's really about my teammates @aashaysanghvi_ & @DivyaDhulipala thank you to all of our frequent ecosystem co-conspirators, our LPs, our co-investors, our founders. i'm very grateful 🙏 and this is the most fun role in the world! semilshah.com/2026/04/26/new…
3
221
Roger Chen :: retweeted
Today we're announcing Sardine's $70M Series C 🚀 Money moves at the speed of light. Crime moves at the speed of AI. Between these forces, trust is eroding faster than ever. We built Sardine to move faster than the threat - and it's working.
42
23
235
91,899
31 Jan 2025
LGND 👀
While we haven't officially launched, can't help but be excited to see this great artcle in the @WSJ featuring @lgnd_ai (lgnd.io) and the opportunity for AI applied to earth observation data. Big stuff ahead! wsj.com/tech/ai/artificial-i…
1
6
725
19 Apr 2024
zero knowledge (zk) has 2013-2015 deep learning vibes
5
1,381
15 Nov 2023
Only because you gave me the opportunity 🙏
15 Nov 2023
Depends on who you’re sending out. @rgrchen found Fetch ($300M exit) and Expanse ($1B exit) during his short time with us.
1
1
24
5,715
14 Oct 2023
sometimes less is more
2
432
14 Oct 2023
we’re so back
3
362
Roger Chen :: retweeted
An important systems bottleneck when working with LLMs is model loading times, but if you get the details right, you can speed up standard implementations by around 20x (over 10 minutes down to around 35 seconds for Llama-2-70B). There are a few bottlenecks to numbers to think through (numbers for a p4de instance) ⏩ S3 -> disk bandwidth (10GB/s) [but it takes a lot of extra work to saturate this] ⏩ disk -> CPU memory (theoretically 16GB/s but practically 5GB/s) ⏩ CPU memory -> GPU memory (16TB/s for PCIe4) Adding these together for a 130GB model gives around 52 seconds S3 -> GPU memory. But you can do much better. In this blog post, we got it down to around 35 seconds (compared to over 10 minutes for standard implementations). Fun fact: you can reverse all of this to build fast model checkpointing (GPU memory -> S3).
How long does it take to download Llama2 70B? On the 4x 25 Gbps NICs that aws.p4de's have, it should take ~10s. Yet in production we've observed much higher times, which makes autoscaling less responsive more expensive. This blog post shows how we've reduced download & init time from 4m25s to 20s, using techniques such as streaming S3->CPU->GPU and multiple TCP streams. anyscale.com/blog/loading-ll… Excellent systems work by @bonecheng!
2
57
374
172,038
Roger Chen :: retweeted
3 Oct 2023
The Rhind Mathematical Papyrus is the oldest manuscript written in algebra and trigonometry, dating back to 3,550 years ago. It shows that the Egyptians used first-order equations, geometric series and a second-order algebraic equation, related to the Pythagorean theorem a² b² = c² It also describes how to obtain an approximation of π accurate to within less than 1% and one of the earliest attempts at squaring the circle.
125
1,772
10,094
2,264,704
18 Nov 2022
the work is the reward
1
2
Roger Chen :: retweeted
One funny thing about DAO governance is that it makes it hard to manage assets held by a collective (why vote to sell the asset you're voting with?) — but for many collectives, asset management is more than pure yield optimization Can we do better with a return to futarchy?
12 Oct 2022
Today we're proud to announce Aera, the world’s first autonomous, data-driven treasury management protocol. Aera will provide DAOs with a one-stop solution for managing their token reserves efficiently and transparently. aera.mirror.xyz/QMSOGh75-_n2…
13
19
124
26 Jun 2022
It’s really great seeing some friends for the first time in a couple years and picking up right where we left off. Thankful for those people.
1
4
11 May 2022
seek credit and you'll find insecurity instead
1
6
6 May 2022
Used to be cool to say you seed invested in a couple unicorns. Now it's like saying you seed invest.
1
3 May 2022
Nurses are a different breed and I’m not sure how we’d have a functioning society without them
25 Apr 2022
overthinking results from the gluttony of time
1
21 Apr 2022
all this shitposting on twitter I might have to start kindposting up in here
3
9 Mar 2022
*Something special* Couldn’t be more excited for some truly amazing friends. Fully convinced they’ll shape a better world in the years to come. Congrats guys :)
8 Mar 2022
SOME NEWS: after a decade investing in the most brilliant interdisciplinary entrepreneurs on the planet, my best friends & partners @nanli @AGoulburn & I are joining forces on … *something new* ;) “If the path be beautiful, let us not ask where it leads.” - Anatole France
1
1
13
22 Feb 2022
22222 02222022
2